Many daily activities present information in the form of a stream of text, and often people can benefit from additional information on the topic discussed. TV broadcast news can be treated as one such stream of text; in this paper we discuss finding news articles on the web that are relevant to news currently being broadcast. We evaluated a variety of algorithms for this problem, looking at the impact of inverse document frequency, stemming, compounds, history, and query length on the relevance and coverage of news articles returned in real time during a broadcast. We also evaluated several postprocessing techniques for improving the precision, including reranking using additional terms, reranking by document similarity, and filtering on document similarity. For the best algorithm, 84–91 % of the articles found were relevant, with at least 64 % of the articles being on the exact topic of the broadcast. In addition, a relevant article was found for at least 70 % of the topics.

In this paper, we outline the relation between Knowledge Management (KM) as an application area on the one hand, and software agents as a basic technology for supporting KM on the other. We start by presenting characteristics of KM which account for some drawbacks of today's -- typically centralized -- technological approaches for KM. We argue that the basic features of agents (social ability, autonomy, re- and proactiveness) can alleviate several of these drawbacks. A classification schema for the description of agent-based KM systems is established, and a couple of example systems are depicted in terms of this schema. The paper concludes with questions which we think research in Agent-mediated Knowledge Management (AMKM) should deal with.

Electronic concept mapping tools empower experts to play an active role in the knowledge capture process, and provide a medium for building richly connected multimedia knowledge models -- sets of linked concept maps and resources about a particular domain. Knowledge models are intended to be used as a means for sharing knowledge among humans, not as carefully-crafted knowledge bases upon which machines will be performing inference. However, users must still confront the questions of what to include in a concept map and which concept maps to include in a knowledge model. This paper describes ongoing research on methods to provide content-based support to users as they extend concept maps by adding concepts and propositions, and as they select topics for new maps. The goal is to provide scaffolding for experts as they build their own concept maps, link their maps to others', and decide how to extend their knowledge models. The paper presents three approaches which start from a concept map under construction and mine related information -- both from prior concept maps, and from the web -- to propose information to aid the user's knowledge capture and knowledge construction. The paper begins with a brief summary of the concept mapping process and the CmapTools concept mapping software. It then presents three types of implemented suggesters, to suggest concepts, propositions, concept maps, and new topics to aid experts using the CmapTools, and describes preliminary experiments to assess their performance. It closes with a discussion of next steps for testing and refining these methods.

Implicit query systems examine a document and automatically conduct searches for the most relevant information. In this paper, we offer three contributions to implicit query research. First, we show how to use query logs from a search engine: by constraining results to commonly issued queries, we can get dramatic improvements. Second, we describe a method for optimizing parameters for an implicit query system, by using logistic regression training. The method is designed to estimate the probability that any particular suggested query is a good one. Third, we show which features beyond standard TF-IDF features are most helpful in our logistic regression model: query frequency information, capitalization information, subject line information, and message length information. Using the optimization method and the additional features, we are able to produce a system with up to 6 times better results on top-1 score than a simple TF-IDF system.

In recent years we have witnessed the wide-spread evolution of support tools that operate in association with the user to accomplish a range of computer-mediated tasks. Two examples of these tools are critics and recommenders. Critics are cooperative tools that observe the user interacting with a computer system and present reasoned opinions about a product under development. Recommender systems are tools that assist users by facilitating access to relevant items. At the same time, defeasible argumentation has evolved as a successful approach in AI to model commonsense qualitative reasoning, with applications in many areas, such as agent theory, knowledge engineering and legal reasoning. This paper presents a novel approach towards the integration of user support systems, such as critics and recommender systems, with a defeasible argumentation framework. The final goal is to enhance practical reasoning capabilities of current user support tools by incorporating argument-based qualitative inference.

In order for intelligent interfaces to provide proactive assistance, they must customize their behavior based on the user's task context. Existing systems often assess context based on a single snapshot of the user's current activities (e.g., examining the content of the document that the user is currently consulting). However, an accurate picture of the user's context may depend not only on this local information, but also on information about the user's behavior over time. This paper discusses work on a recommender system, Calvin, which learns to identify broader contexts by relating documents that tend to be accessed together. Calvin's text analysis algorithm, WordSieve, develops term vector descriptions of these contexts in real time, without needing to accumulate comprehensive statistics about an entire corpus. Calvin uses these descriptions (1) to index documents to suggest them in similar future contexts and (2) to formulate contextbased queries for search engines. Results of initial experiments are encouraging for the approach's improved ability to associate documents with the research tasks in which they were consulted, compared to methods using only local information. This paper sketches the project goals, the current implementation of the system, and plans for its continued development and evaluation.

Proactive retrieval systems monitor a user's task context and automatically provide the user with related resources. The effectiveness of such systems depends on their ability to perform context-based retrieval, generating queries which return context-relevant results. Two factors make this task especially challenging for Web-based retrieval. First, the quality of Web retrieval can be strongly affected by the vocabulary used to generate the queries. If the system's vocabulary for describing the context differs from the vocabulary used in the resources themselves, relevant resources may be missed. Second, search engine restrictions on query length may make it difficult to include sufficient contextual information in a single query. This paper presents an algorithm, IACS (Incremental Algorithm for Context-Based Search), which addresses these problems by building up, applying, and refining partial context descriptions incrementally. In IACS, an initial term-based context description is the starting point for a cycle of mining search engines, performing context-based filtering of results, and refining context descriptions to generate new rounds of queries in an expanded vocabulary. IACS has been applied in a system for proactively supporting concept-map-based knowledge modeling, by retrieving resources relevant to target concepts in the context of the rich information provided by "in progress" concept maps. An evaluation of the system shows that it provides significant improvements over a baseline for retrieving context-relevant resources. We expect the algorithm to have broad applicability to context-based Web retrieval for rich contexts.

Recommender systems have evolved in the last years as specialized tools to assist users in a plethora of computermediated tasks by providing guidelines or hints. Most recommender systems are aimed at facilitating access to relevant items, a situation particularly common when performing web-based tasks. At the same time, defeasible argumentation has evolved as a successful approach in AI to model commonsense qualitative reasoning, with applications in many areas, such as agent theory, knowledge engineering and legal reasoning. This paper presents a first approach towards the integration of web-based recommender systems with a defeasible argumentation framework. The final goal is to enhance practical reasoning capabilities of current recommender system technology by incorporating argument-based qualitative inference.

Our project is aimed at integrating and extending inferential, analogical, and intelligent IR technologies to create power tools for intelligence analysts. Our goal is to discover interesting and powerful functional integrations that permit these technologies to exploit each others ’ strengths to mitigate their weaknesses. From the perspective of knowledge-based AI technology, a key goal of the project is to extend the reach of such systems into the world of unstructured data and text. From the perspective of IR technology, it is to leverage the application of inferential and analogical techniques to structured representations in order to achieve significant new functionality. Background Intelligence analysts must sift through massive amounts of data, using perspective gained from history and experience to pull together from disparate sources the best coherent picture of what is happening. Intelligent information technology has the potential to create new software tools that could aid analysts in a number of critical and mutually reinforcing ways: Analysts make heavy use of precedents and analogies. This sometimes leads to vital “trans-logical ” leaps. Because of fundamental human cognitive limitations, it also sometimes leads to false analogies, where the matches are too superficial, and to missed opportunities, where the matches are too obscure for unaided human reasoning to uncover. A suite of knowledge-based software power tools could help analysts recognize deeper or less obvious analogies, could help them apply these analogies to the current situation, and could help them reject superficially plausible but useless analogies and precedents more rapidly. Analysts make heavy use of scenario generation, both to interpret reported data and to project plausible future events. Again because of fundamental human cognitive limitations, the first one or two plausible interpretations (or

Abstract. The success of personal information agents depends on their capacity to both identify relevant information for users and proactively recommend context-relevant information. In this paper, we propose an approach to enable proactive context-aware recommendation based on the knowledge of both user interests and browsing patterns. The proposed approach analyzes the browsing behavior of users to derive a semantically enhanced context that points out the information which is likely to be relevant for a user according to its current activities. 1